Large Language Models and the Varieties of Meaning

Abstract

This article responds to Steffen Koch’s, ‘Babbling stochastic parrots? A Kripkean argument for reference in large language models’ (this journal). Koch attempts by appealing to the causal theory of reference to show that when LLM’s use (or seem to use) proper names, they refer to what those names standardly refer to in the language from which they were drawn; Koch then suggests ways to generalize the strategy beyond proper names to kind terms, both natural and social. We argue here that Koch’s argumentative strategy is needlessly controversial and even if successful, would be of limited scope. We offer instead a more direct argument for the conclusion that LLM’s use language (rather than merely appearing to do so) because they meet the conditions required to perform rudimentary acts that Austin (1975) termed phatic acts. The argument applies to language generally, and implies that when so used, that language carries its usual linguistic meaning.